import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import os
import zipfile
import random
from PIL import Image
from typing import List
from datetime import datetime

# For Data Visualization
import seaborn as sns

# For Model Building
import tensorflow as tf
import keras
from tensorflow.keras.models import Sequential, Model  # Sequential API for sequential model
from tensorflow.keras.layers import Dense, Dropout, Flatten, Input  # Importing different layers
from tensorflow.keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D, BatchNormalization, LeakyReLU, ReLU
from tensorflow.keras import backend
from tensorflow.keras.optimizers import Adam, RMSprop  # Optimizers for optimizing the model
from tensorflow.keras.callbacks import EarlyStopping  # Regularization method to prevent the overfitting
from tensorflow.keras.callbacks import ModelCheckpoint, ReduceLROnPlateau
from tensorflow.keras.preprocessing.image import load_img, ImageDataGenerator
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from keras.applications.vgg16 import preprocess_input as preprocess_input_vgg16
from keras.applications import VGG16
from keras.applications.resnet_v2 import preprocess_input as preprocess_input_resnetv2
from keras.applications import ResNet50V2
from keras.applications.efficientnet_v2 import preprocess_input as preprocess_input_efficientnetv2
from keras.applications import EfficientNetV2B0

2024-04-11 00:04:06.669124: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.

2024-04-11 00:04:07.217980: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

# Storing the path of the data file from the Google drive
path = "/home/iamtxena/sandbox/mit-ai/capstone/Facial_Emotion_Recognition/Facial_emotion_images.zip"

# The data is provided as a zip file so we need to extract the files from the zip file
with zipfile.ZipFile(path, "r") as zip_ref:
    zip_ref.extractall()

DATADIR = "/home/iamtxena/sandbox/mit-ai/capstone/Facial_Emotion_Recognition/Facial_emotion_images"  # Base directory
SUBDIRS = ["train", "validation", "test"]  # Subdirectories
CATEGORIES = ["happy", "neutral", "sad", "surprise"]  # Emotion categories

def get_first_image_size(directory, sub_dirs, categories):
    """
    Returns the size of the first image found in the specified directories.

    Parameters:
    - directory (str): The base directory of the dataset.
    - sub_dirs (list of str): Subdirectories to search through (e.g., ['train', 'validation', 'test']).
    - categories (list of str): Categories (e.g., ['happy', 'neutral', 'sad', 'surprise']).

    Returns:
    - tuple: Size of the first image found (width, height).
    """
    for sub_dir in sub_dirs:
        for category in categories:
            path = os.path.join(directory, sub_dir, category)
            for img_name in os.listdir(path):
                img_path = os.path.join(path, img_name)
                with Image.open(img_path) as img:
                    return img.size  # Return the size of the first image found

# Get the size of the first image
expected_size = get_first_image_size(DATADIR, SUBDIRS, CATEGORIES)

print(f"Expected size of the first image: {expected_size}")

Expected size of the first image: (48, 48)

def check_image_sizes(directory, sub_dirs, categories, target_size):
    """
    Checks if all images in the specified directories match the target size.

    Parameters:
    - directory (str): The base directory of the dataset.
    - sub_dirs (list of str): Subdirectories to search through.
    - categories (list of str): Categories to search through.
    - target_size (tuple): The expected size of the images (width, height).

    Returns:
    - bool: True if all images match the target size, False otherwise.
    """
    all_match = True  # Flag to keep track of size match

    for sub_dir in sub_dirs:
        for category in categories:
            path = os.path.join(directory, sub_dir, category)
            for img_name in os.listdir(path):
                img_path = os.path.join(path, img_name)
                with Image.open(img_path) as img:
                    if img.size != target_size:
                        print(f"Image {img_path} has a different size: {img.size}, expected: {target_size}")
                        all_match = False
                        return all_match  # Return early upon first mismatch
    return all_match

# Check if all images match the expected size
all_match = check_image_sizes(DATADIR, SUBDIRS, CATEGORIES, expected_size)
if all_match:
    print("All images match the expected size.")
else:
    print("Not all images match the expected size.")

All images match the expected size.

def visualize_emotion_images(directory: str, sub_dirs: List[str], emotion: str, image_count: int = 9) -> None:
    """
    Visualizes a specified number of images from a given emotion class directory across specified subdirectories.

    Parameters:
    - directory (str): The base directory where emotion class folders are located across subdirectories.
    - sub_dirs (List[str]): List of subdirectories ('train', 'validation', 'test') to search through.
    - emotion (str): The specific emotion class to visualize images from.
    - image_count (int): The number of images to display. Defaults to 9.

    Returns:
    - None: This function does not return any value but displays images inline.
    """
    image_paths: List[str] = []  # To store paths of images to be displayed

    # Iterate through the specified subdirectories to collect image paths
    for sub_dir in sub_dirs:
        emotion_dir: str = os.path.join(directory, sub_dir, emotion)
        if os.path.isdir(emotion_dir):
            for img_name in os.listdir(emotion_dir):
                img_path = os.path.join(emotion_dir, img_name)
                image_paths.append(img_path)

    # If there are no images found for the emotion, print a message and return
    if not image_paths:
        print(f"No images found for the specified emotion: {emotion}")
        return

    # Select a random subset of image paths
    selected_image_paths: np.ndarray = np.random.choice(image_paths, min(image_count, len(image_paths)), replace=False)

    # Setup for plotting
    fig = plt.figure(figsize=(4, 4))
    columns: int = 3
    rows: int = image_count // columns + (1 if image_count % columns else 0)

    # Iterate over the selected images and display them
    for i, image_path in enumerate(selected_image_paths, start=1):
        ax = fig.add_subplot(rows, columns, i)
        image = load_img(image_path, target_size=(48, 48))  # Ensure the image is resized to 48x48
        plt.imshow(image)
        plt.axis("off")

    plt.tight_layout()
    plt.show()

visualize_emotion_images(DATADIR, SUBDIRS, "happy", 9)

visualize_emotion_images(DATADIR, SUBDIRS, "sad", 9)

visualize_emotion_images(DATADIR, SUBDIRS, "neutral", 9)

visualize_emotion_images(DATADIR, SUBDIRS, "surprise", 9)

# Function to count images in each category
def count_images(data_dir, categories):
    counts = []
    for category in categories:
        path = os.path.join(data_dir, category)
        count = len([name for name in os.listdir(path) if os.path.isfile(os.path.join(path, name))])
        counts.append(count)
    return counts


SUBDIRS_DICT = {"train": "train", "validation": "validation", "test": "test"}

# Counting images in each dataset
train_counts = count_images(os.path.join(DATADIR, SUBDIRS_DICT["train"]), CATEGORIES)
validation_counts = count_images(os.path.join(DATADIR, SUBDIRS_DICT["validation"]), CATEGORIES)
test_counts = count_images(os.path.join(DATADIR, SUBDIRS_DICT["test"]), CATEGORIES)


# Create DataFrames and format for easier reading
def create_df(counts, categories, dataset_name):
    df = pd.DataFrame({"Class": categories, "Count": counts})
    df["Percentage"] = (df["Count"] / df["Count"].sum()) * 100
    df.set_index("Class", inplace=True)

    # Formatting for easier reading
    df["Count"] = df["Count"].apply(lambda x: f"{x:,}")  # Adds commas to thousands
    df["Percentage"] = df["Percentage"].apply(lambda x: f"{x:.2f}")  # Rounds to two decimals

    print(f"{dataset_name} Data Distribution:")
    print(df)
    total_images = df["Count"].str.replace(",", "").astype(int).sum()
    print(f"Total images in {dataset_name}: {total_images:,}\n")  # Formats total count with commas


create_df(train_counts, CATEGORIES, "Training")
create_df(validation_counts, CATEGORIES, "Validation")
create_df(test_counts, CATEGORIES, "Testing")

Training Data Distribution:
          Count Percentage
Class                     
happy     3,976      26.32
neutral   3,978      26.33
sad       3,982      26.36
surprise  3,173      21.00
Total images in Training: 15,109

Validation Data Distribution:
          Count Percentage
Class                     
happy     1,825      36.67
neutral   1,216      24.43
sad       1,139      22.89
surprise    797      16.01
Total images in Validation: 4,977

Testing Data Distribution:
         Count Percentage
Class                    
happy       32      25.00
neutral     32      25.00
sad         32      25.00
surprise    32      25.00
Total images in Testing: 128

# Set this to 'grayscale' as the images are in grayscale
color_mode = "grayscale"

# As we have checked, all images are 48x48, we will set the img_width and img_height to 48
img_width, img_height = 48, 48
color_layers = 1
# A batch size of 32 is appropriate for this dataset provide to provide a good balance
# between the model's ability to generalize (avoid overfitting) and computational efficiency.
batch_size = 32

# Training Data Augmentation
train_datagen = ImageDataGenerator(
    rescale=1.0 / 255,  # Normalize pixel values to [0,1]
    horizontal_flip=True,  # Faces are symmetric; flipping can simulate looking from another direction
    brightness_range=(0.5, 1.5),  # Randomly adjust brightness to simulate different lighting conditions
    shear_range=0.3,  # Shear transformations for perspective changes
    rotation_range=20,  # Slight rotation to introduce variability without distorting emotion features
    width_shift_range=0.1,  # Slight horizontal shifts to simulate off-center faces
    height_shift_range=0.1,  # Slight vertical shifts to account for different heights/angles
    zoom_range=0.1,  # Small zoom in/out to simulate closer or further away faces
)

# Validation and Testing Data should not be augmented!
validation_datagen = ImageDataGenerator(rescale=1.0 / 255)
test_datagen = ImageDataGenerator(rescale=1.0 / 255)

# Creating train_dir, validation_dir, and test_dir directories using the structure of DATADIR and SUBDIRS
train_dir = os.path.join(DATADIR, SUBDIRS_DICT["train"])
validation_dir = os.path.join(DATADIR, SUBDIRS_DICT["validation"])
test_dir = os.path.join(DATADIR, SUBDIRS_DICT["test"])

# Train Generator
train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
)

# Validation Generator
validation_generator = validation_datagen.flow_from_directory(
    validation_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for evaluation
)

# Testing Generator
test_generator = test_datagen.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for testing
)

Found 15109 images belonging to 4 classes.

Found 4977 images belonging to 4 classes.

Found 128 images belonging to 4 classes.

# Fetch a batch of images and labels
images, labels = next(train_generator)

# Assuming the labels are one-hot encoded, we need to convert them back to class indices
labels_indices = labels.argmax(axis=1)

# Mapping of indices to class names, based on the 'class_indices' attribute of the generator
index_to_class = {v: k for k, v in train_generator.class_indices.items()}

fig, axes = plt.subplots(4, 4, figsize=(8, 8))
for image, label_index, ax in zip(images, labels_indices, axes.flatten()):
    ax.imshow(image.squeeze(), cmap="gray")  # Squeeze and cmap for grayscale
    class_name = index_to_class[label_index]
    ax.set_title(class_name)
    ax.axis("off")

plt.tight_layout()
plt.show()

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

# Intializing a sequential model
model_1 = Sequential()

model_1.add(Input(shape=(img_width, img_height, color_layers)))

model_1.add(Conv2D(64, (3, 3), activation="relu", padding="same"))
model_1.add(MaxPooling2D((2, 2), padding="same"))
model_1.add(Dropout(0.2))

# Adding second conv layer with 32 filters
model_1.add(Conv2D(32, (3, 3), activation="relu", padding="same"))
model_1.add(MaxPooling2D((2, 2), padding="same"))
model_1.add(Dropout(0.2))

# Add third conv layer with 32 filters and kernel size 3x3, padding 'same' followed by a Maxpooling2D layer
model_1.add(Conv2D(32, (3, 3), activation="relu", padding="same"))
model_1.add(MaxPooling2D((2, 2), padding="same"))
model_1.add(Dropout(0.2))

# Flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model_1.add(Flatten())

# Adding fully connected dense layers
model_1.add(Dense(512, activation="relu"))
model_1.add(Dense(64, activation="relu"))

# Adding output layer
model_1.add(Dense(4, activation="softmax"))

# Using Adam Optimizer
opt = Adam(learning_rate=0.001)

2024-04-11 00:04:11.726218: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.747212: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.747375: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.747960: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.748070: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.748171: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.802454: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.802581: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.802685: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-04-11 00:04:11.802760: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1928] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 6272 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3070, pci bus id: 0000:06:00.0, compute capability: 8.6

# Compiling the model
model_1.compile(optimizer=opt, loss="categorical_crossentropy", metrics=["accuracy"])

# Generating the summary of the model
model_1.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ conv2d (Conv2D)                 │ (None, 48, 48, 64)     │           640 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d (MaxPooling2D)    │ (None, 24, 24, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout (Dropout)               │ (None, 24, 24, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_1 (Conv2D)               │ (None, 24, 24, 32)     │        18,464 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_1 (MaxPooling2D)  │ (None, 12, 12, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_1 (Dropout)             │ (None, 12, 12, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_2 (Conv2D)               │ (None, 12, 12, 32)     │         9,248 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_2 (MaxPooling2D)  │ (None, 6, 6, 32)       │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_2 (Dropout)             │ (None, 6, 6, 32)       │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten (Flatten)               │ (None, 1152)           │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ (None, 512)            │       590,336 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 64)             │        32,832 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ (None, 4)              │           260 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 651,780 (2.49 MB)

 Trainable params: 651,780 (2.49 MB)

 Non-trainable params: 0 (0.00 B)

results_path = "/home/iamtxena/sandbox/mit-ai/capstone/Facial_Emotion_Recognition/results"

class DelayedEarlyStopping(EarlyStopping):
    """Stop training when a monitored metric has stopped improving after a certain number of epochs.

    Arguments:
        monitor: Quantity to be monitored.
        min_delta: Minimum change in the monitored quantity to qualify as an improvement,
                   i.e., an absolute change of less than min_delta will count as no improvement.
        patience: Number of epochs with no improvement after which training will be stopped.
        verbose: Verbosity mode.
        mode: One of `{'auto', 'min', 'max'}`. In `min` mode, training will stop when the
              quantity monitored has stopped decreasing; in `max` mode it will stop when the
              quantity monitored has stopped increasing; in `auto` mode, the direction is
              automatically inferred from the name of the monitored quantity.
        baseline: Baseline value for the monitored quantity. Training will stop if the model
                  doesn't show improvement over the baseline.
        restore_best_weights: Whether to restore model weights from the epoch with the best value
                              of the monitored quantity.
        start_epoch: The epoch on which to start considering early stopping. Before this epoch,
                     early stopping will not be considered. This ensures that early stopping
                     checks only after a certain number of epochs.
    """

    def __init__(
        self,
        monitor="val_loss",
        min_delta=0,
        patience=0,
        verbose=0,
        mode="auto",
        baseline=None,
        restore_best_weights=False,
        start_epoch=30,
    ):
        super().__init__(
            monitor=monitor,
            min_delta=min_delta,
            patience=patience,
            verbose=verbose,
            mode=mode,
            baseline=baseline,
            restore_best_weights=restore_best_weights,
        )
        self.start_epoch = start_epoch

    def on_epoch_end(self, epoch, logs=None):
        # Override the original `on_epoch_end` method to include `start_epoch` logic.

        # If the current epoch is less than the start epoch, skip the early stopping check
        if epoch < self.start_epoch:
            return

        # Call the parent class method to perform the regular early stopping checks after the start epoch
        super().on_epoch_end(epoch, logs)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 30 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=30
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

# Define the saving the best model callback
mc = ModelCheckpoint(
    f"{results_path}/best_model_1_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 45 epochs and using validation set
history_1 = model_1.fit(
    train_generator,
    epochs=45,
    validation_data=validation_generator,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/45

/home/iamtxena/sandbox/mit-ai/my_env/lib/python3.10/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:120: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1712793853.651159 1474977 service.cc:145] XLA service 0x7f226c00e760 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1712793853.651180 1474977 service.cc:153]   StreamExecutor device (0): NVIDIA GeForce RTX 3070, Compute Capability 8.6
2024-04-11 00:04:13.681751: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.

2024-04-11 00:04:13.811989: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:465] Loaded cuDNN version 8907

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1712793855.129443 1475073 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_1953', 4 bytes spill stores, 4 bytes spill loads

  1/473 ━━━━━━━━━━━━━━━━━━━━ 37:55 5s/step - accuracy: 0.0938 - loss: 1.3994


  6/473 ━━━━━━━━━━━━━━━━━━━━ 4s 10ms/step - accuracy: 0.2084 - loss: 1.3991


  9/473 ━━━━━━━━━━━━━━━━━━━━ 5s 13ms/step - accuracy: 0.2167 - loss: 1.4005


 13/473 ━━━━━━━━━━━━━━━━━━━━ 5s 13ms/step - accuracy: 0.2237 - loss: 1.3999

I0000 00:00:1712793857.604780 1474977 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.


 17/473 ━━━━━━━━━━━━━━━━━━━━ 5s 13ms/step - accuracy: 0.2293 - loss: 1.3987

plt.plot(history_1.history["accuracy"])
plt.plot(history_1.history["val_accuracy"])
plt.title("CNN Model 1 accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = model_1.evaluate(test_generator, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 21ms/step - accuracy: 0.8438 - loss: 0.4053


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - accuracy: 0.7573 - loss: 0.5913

Loss: 0.65003901720047, Accuracy: 0.734375

pred_probabilities = model_1.predict(test_generator, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("CNN Model 1 Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 1s 365ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step

              precision    recall  f1-score   support

       happy       0.84      0.84      0.84        32
     neutral       0.65      0.69      0.67        32
         sad       0.54      0.59      0.57        32
    surprise       0.96      0.81      0.88        32

    accuracy                           0.73       128
   macro avg       0.75      0.73      0.74       128
weighted avg       0.75      0.73      0.74       128

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

# Initializing a sequential model
model_2 = Sequential()

model_2.add(Input(shape=(img_width, img_height, color_layers)))

# First Convolutional Block
model_2.add(Conv2D(256, kernel_size=2, padding="same"))
model_2.add(BatchNormalization())
model_2.add(LeakyReLU(negative_slope=0.1))
model_2.add(MaxPooling2D(pool_size=2))

# Second Convolutional Block
model_2.add(Conv2D(128, kernel_size=2, padding="same"))
model_2.add(BatchNormalization())
model_2.add(LeakyReLU(negative_slope=0.1))
model_2.add(MaxPooling2D(pool_size=2))

# Third Convolutional Block
model_2.add(Conv2D(64, kernel_size=2, padding="same"))
model_2.add(BatchNormalization())
model_2.add(LeakyReLU(negative_slope=0.1))
model_2.add(MaxPooling2D(pool_size=2))

# Fourth Convolutional Block
model_2.add(Conv2D(32, kernel_size=2, padding="same"))
model_2.add(BatchNormalization())
model_2.add(LeakyReLU(negative_slope=0.1))
model_2.add(MaxPooling2D(pool_size=2))

# Flatten the output of the conv layers to feed into the dense layers
model_2.add(Flatten())

# Fully connected layers
model_2.add(Dense(512, activation="relu"))
model_2.add(Dense(128, activation="relu"))
model_2.add(Dense(64))
model_2.add(BatchNormalization())
model_2.add(ReLU())  # Using ReLU after batch normalization

# Adding output layer
model_2.add(Dense(4, activation="softmax"))

# Using Adam Optimizer
optimizer = Adam(learning_rate=0.001)

# Compile the model
model_2.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"])

model_2.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ conv2d (Conv2D)                 │ (None, 48, 48, 256)    │         1,280 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization             │ (None, 48, 48, 256)    │         1,024 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu (LeakyReLU)         │ (None, 48, 48, 256)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d (MaxPooling2D)    │ (None, 24, 24, 256)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_1 (Conv2D)               │ (None, 24, 24, 128)    │       131,200 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_1           │ (None, 24, 24, 128)    │           512 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_1 (LeakyReLU)       │ (None, 24, 24, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_1 (MaxPooling2D)  │ (None, 12, 12, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_2 (Conv2D)               │ (None, 12, 12, 64)     │        32,832 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_2           │ (None, 12, 12, 64)     │           256 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_2 (LeakyReLU)       │ (None, 12, 12, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_2 (MaxPooling2D)  │ (None, 6, 6, 64)       │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_3 (Conv2D)               │ (None, 6, 6, 32)       │         8,224 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_3           │ (None, 6, 6, 32)       │           128 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_3 (LeakyReLU)       │ (None, 6, 6, 32)       │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_3 (MaxPooling2D)  │ (None, 3, 3, 32)       │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten (Flatten)               │ (None, 288)            │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ (None, 512)            │       147,968 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 128)            │        65,664 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ (None, 64)             │         8,256 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_4           │ (None, 64)             │           256 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu (ReLU)                    │ (None, 64)             │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_3 (Dense)                 │ (None, 4)              │           260 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 397,860 (1.52 MB)

 Trainable params: 396,772 (1.51 MB)

 Non-trainable params: 1,088 (4.25 KB)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 30 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=30
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

# Define the saving the best model callback
mc = ModelCheckpoint(
    f"{results_path}/best_model_2_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 45 epochs and using validation set
history_2 = model_2.fit(
    train_generator,
    epochs=45,
    validation_data=validation_generator,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/45

I0000 00:00:1712794150.228648 1482664 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_2631', 4 bytes spill stores, 4 bytes spill loads

  1/473 ━━━━━━━━━━━━━━━━━━━━ 42:28 5s/step - accuracy: 0.3438 - loss: 1.3840


  5/473 ━━━━━━━━━━━━━━━━━━━━ 7s 17ms/step - accuracy: 0.2798 - loss: 1.6040


  8/473 ━━━━━━━━━━━━━━━━━━━━ 7s 17ms/step - accuracy: 0.2899 - loss: 1.6063


 13/473 ━━━━━━━━━━━━━━━━━━━━ 6s 15ms/step - accuracy: 0.2964 - loss: 1.6224


 18/473 ━━━━━━━━━━━━━━━━━━━━ 6s 14ms/step - accuracy: 0.2983 - loss: 1.6201


 22/473 ━━━━━━━━━━━━━━━━━━━━ 6s 14ms/step - accuracy: 0.2994 - loss: 1.6173


 26/473 ━━━━━━━━━━━━━━━━━━━━ 6s 14ms/step - accuracy: 0.3013 - loss: 1.6110


 31/473 ━━━━━━━━━━━━━━━━━━━━ 6s 14ms/step - accuracy: 0.3042 - loss: 1.6004

plt.plot(history_2.history["accuracy"])
plt.plot(history_2.history["val_accuracy"])
plt.title("CNN Model 2 accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = model_2.evaluate(test_generator, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - accuracy: 0.8125 - loss: 0.4481


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - accuracy: 0.7875 - loss: 0.5252

Loss: 0.5420138835906982, Accuracy: 0.7890625

pred_probabilities = model_2.predict(test_generator, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("CNN Model 2 Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 1s 394ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step

              precision    recall  f1-score   support

       happy       0.81      0.81      0.81        32
     neutral       0.66      0.78      0.71        32
         sad       0.75      0.66      0.70        32
    surprise       0.97      0.91      0.94        32

    accuracy                           0.79       128
   macro avg       0.80      0.79      0.79       128
weighted avg       0.80      0.79      0.79       128

# Set this to 'rgb' as this is the required format for the transfer learning architectures
color_mode = "rgb"
color_layers = 3
# Using the same size as before for the images
img_width, img_height = 48, 48
# A batch size of 32 is appropriate for this dataset provide to provide a good balance
# between the model's ability to generalize (avoid overfitting) and computational efficiency.
batch_size = 32

# Training Data Augmentation for VGG16
train_datagen_vgg16 = ImageDataGenerator(
    preprocessing_function=preprocess_input_vgg16,  # Use model-specific preprocessing
    horizontal_flip=True,  # Faces are symmetric; flipping can simulate looking from another direction
    brightness_range=(0.5, 1.5),  # Randomly adjust brightness to simulate different lighting conditions
    shear_range=0.3,  # Shear transformations for perspective changes
    rotation_range=20,  # Slight rotation to introduce variability without distorting emotion features
    width_shift_range=0.1,  # Slight horizontal shifts to simulate off-center faces
    height_shift_range=0.1,  # Slight vertical shifts to account for different heights/angles
    zoom_range=0.1,  # Small zoom in/out to simulate closer or further away faces
)

# Training Data Augmentation for ResNet
train_datagen_resnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_resnetv2,  # Use model-specific preprocessing
    horizontal_flip=True,  # Faces are symmetric; flipping can simulate looking from another direction
    brightness_range=(0.5, 1.5),  # Randomly adjust brightness to simulate different lighting conditions
    shear_range=0.3,  # Shear transformations for perspective changes
    rotation_range=20,  # Slight rotation to introduce variability without distorting emotion features
    width_shift_range=0.1,  # Slight horizontal shifts to simulate off-center faces
    height_shift_range=0.1,  # Slight vertical shifts to account for different heights/angles
    zoom_range=0.1,  # Small zoom in/out to simulate closer or further away faces
)

# Training Data Augmentation for EfficientNet
train_datagen_efficientnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_efficientnetv2,  # Use model-specific preprocessing
    horizontal_flip=True,  # Faces are symmetric; flipping can simulate looking from another direction
    brightness_range=(0.5, 1.5),  # Randomly adjust brightness to simulate different lighting conditions
    shear_range=0.3,  # Shear transformations for perspective changes
    rotation_range=20,  # Slight rotation to introduce variability without distorting emotion features
    width_shift_range=0.1,  # Slight horizontal shifts to simulate off-center faces
    height_shift_range=0.1,  # Slight vertical shifts to account for different heights/angles
    zoom_range=0.1,  # Small zoom in/out to simulate closer or further away faces
)

# Validation and Testing Data should not be augmented! VGG16 version
validation_datagen_vgg16 = ImageDataGenerator(
    preprocessing_function=preprocess_input_vgg16
)  # Use model-specific preprocessing
test_datagen_vgg16 = ImageDataGenerator(
    preprocessing_function=preprocess_input_vgg16
)  # Use model-specific preprocessing

# Validation and Testing Data should not be augmented! ResNet version
validation_datagen_resnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_resnetv2
)  # Use model-specific preprocessing
test_datagen_resnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_resnetv2
)  # Use model-specific preprocessing

# Validation and Testing Data should not be augmented! Efficient Net version
validation_datagen_efficientnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_efficientnetv2
)  # Use model-specific preprocessing
test_datagen_efficientnet = ImageDataGenerator(
    preprocessing_function=preprocess_input_efficientnetv2
)  # Use model-specific preprocessing


# Creating train_dir, validation_dir, and test_dir with the structure of DATADIR and SUBDIRS
train_dir = os.path.join(DATADIR, SUBDIRS_DICT["train"])
validation_dir = os.path.join(DATADIR, SUBDIRS_DICT["validation"])
test_dir = os.path.join(DATADIR, SUBDIRS_DICT["test"])

# Train Generator VGG16
train_generator_vgg16 = train_datagen_vgg16.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
)

# Train Generator ResNet
train_generator_resnet = train_datagen_resnet.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
)

# Train Generator EfficientNet
train_generator_efficientnet = train_datagen_efficientnet.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
)

# Validation Generator VGG16
validation_generator_vgg16 = validation_datagen_vgg16.flow_from_directory(
    validation_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for evaluation
)

# Validation Generator ResNet
validation_generator_resnet = validation_datagen_resnet.flow_from_directory(
    validation_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for evaluation
)

# Validation Generator EfficientNet
validation_generator_efficientnet = validation_datagen_efficientnet.flow_from_directory(
    validation_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for evaluation
)

# Testing Generator VGG16
test_generator_vgg16 = test_datagen_vgg16.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for testing
)

# Testing Generator ResNet
test_generator_resnet = test_datagen_resnet.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for testing
)

# Testing Generator EfficientNet
test_generator_efficientnet = test_datagen_efficientnet.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for testing
)

Found 15109 images belonging to 4 classes.

Found 15109 images belonging to 4 classes.

Found 15109 images belonging to 4 classes.

Found 4977 images belonging to 4 classes.

Found 4977 images belonging to 4 classes.

Found 4977 images belonging to 4 classes.

Found 128 images belonging to 4 classes.

Found 128 images belonging to 4 classes.

Found 128 images belonging to 4 classes.

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

vgg_model = VGG16(weights="imagenet", include_top=False, input_shape=(img_width, img_height, color_layers))
vgg_model.summary()

Model: "vgg16"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer (InputLayer)        │ (None, 48, 48, 3)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block1_conv1 (Conv2D)           │ (None, 48, 48, 64)     │         1,792 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block1_conv2 (Conv2D)           │ (None, 48, 48, 64)     │        36,928 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block1_pool (MaxPooling2D)      │ (None, 24, 24, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block2_conv1 (Conv2D)           │ (None, 24, 24, 128)    │        73,856 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block2_conv2 (Conv2D)           │ (None, 24, 24, 128)    │       147,584 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block2_pool (MaxPooling2D)      │ (None, 12, 12, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block3_conv1 (Conv2D)           │ (None, 12, 12, 256)    │       295,168 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block3_conv2 (Conv2D)           │ (None, 12, 12, 256)    │       590,080 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block3_conv3 (Conv2D)           │ (None, 12, 12, 256)    │       590,080 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block3_pool (MaxPooling2D)      │ (None, 6, 6, 256)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block4_conv1 (Conv2D)           │ (None, 6, 6, 512)      │     1,180,160 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block4_conv2 (Conv2D)           │ (None, 6, 6, 512)      │     2,359,808 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block4_conv3 (Conv2D)           │ (None, 6, 6, 512)      │     2,359,808 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block4_pool (MaxPooling2D)      │ (None, 3, 3, 512)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block5_conv1 (Conv2D)           │ (None, 3, 3, 512)      │     2,359,808 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block5_conv2 (Conv2D)           │ (None, 3, 3, 512)      │     2,359,808 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block5_conv3 (Conv2D)           │ (None, 3, 3, 512)      │     2,359,808 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ block5_pool (MaxPooling2D)      │ (None, 1, 1, 512)      │             0 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 14,714,688 (56.13 MB)

 Trainable params: 14,714,688 (56.13 MB)

 Non-trainable params: 0 (0.00 B)

# Define a new model that cuts VGG16 at the 'block3_pool' layer
model_output = vgg_model.get_layer("block3_pool").output
cut_model = Model(inputs=vgg_model.input, outputs=model_output)

for layer in vgg_model.layers:
    layer.trainable = False

new_vgg16_model = Sequential()

# Adding the convolutional part of the VGG16 model from above
new_vgg16_model.add(cut_model)

# Reduces each feature map to a single value by averaging all elements
new_vgg16_model.add(GlobalAveragePooling2D())

# Adding full connected layers
new_vgg16_model.add(Dense(512, activation="relu"))
new_vgg16_model.add(Dense(128, activation="relu"))
new_vgg16_model.add(Dense(64))
new_vgg16_model.add(BatchNormalization())
new_vgg16_model.add(ReLU())  # Using ReLU after batch normalization

# Adding output layer
new_vgg16_model.add(Dense(4, activation="softmax"))

# Using RMSprop Optimizer
optimizer = RMSprop(learning_rate=0.001)

new_vgg16_model.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"])
new_vgg16_model.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ functional_1 (Functional)       │ ?                      │     1,735,488 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ global_average_pooling2d        │ ?                      │   0 (unbuilt) │
│ (GlobalAveragePooling2D)        │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization             │ ?                      │   0 (unbuilt) │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu (ReLU)                    │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_3 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 1,735,488 (6.62 MB)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 1,735,488 (6.62 MB)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 20 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=20
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

# Define the saving the best model callback
mc = ModelCheckpoint(
    f"{results_path}/best_model_vgg16_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 40 epochs and using validation set
history_vgg = new_vgg16_model.fit(
    train_generator_vgg16,
    epochs=40,
    validation_data=validation_generator_vgg16,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/40

/home/iamtxena/sandbox/mit-ai/my_env/lib/python3.10/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:120: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()

I0000 00:00:1712794452.789847 1490572 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_1155', 40 bytes spill stores, 40 bytes spill loads

I0000 00:00:1712794452.792128 1490580 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_1155', 4 bytes spill stores, 4 bytes spill loads

I0000 00:00:1712794453.017261 1490573 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_473', 8 bytes spill stores, 8 bytes spill loads

I0000 00:00:1712794453.113233 1490576 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_1155', 32 bytes spill stores, 32 bytes spill loads

  1/473 ━━━━━━━━━━━━━━━━━━━━ 21:23 3s/step - accuracy: 0.3750 - loss: 1.4510


  5/473 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.3528 - loss: 1.4957


  7/473 ━━━━━━━━━━━━━━━━━━━━ 9s 19ms/step - accuracy: 0.3495 - loss: 1.4843


 10/473 ━━━━━━━━━━━━━━━━━━━━ 9s 20ms/step - accuracy: 0.3433 - loss: 1.4681


 13/473 ━━━━━━━━━━━━━━━━━━━━ 9s 20ms/step - accuracy: 0.3409 - loss: 1.4549


 16/473 ━━━━━━━━━━━━━━━━━━━━ 8s 20ms/step - accuracy: 0.3435 - loss: 1.4404

plt.plot(history_vgg.history["accuracy"])
plt.plot(history_vgg.history["val_accuracy"])
plt.title("VGG16 Model accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator_vgg16.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = new_vgg16_model.evaluate(test_generator_vgg16, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 21ms/step - accuracy: 0.8750 - loss: 0.5965


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - accuracy: 0.6979 - loss: 0.7874

Loss: 0.8395289778709412, Accuracy: 0.65625

pred_probabilities = new_vgg16_model.predict(test_generator_vgg16, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator_vgg16.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("VGG16 Model Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 1s 461ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step

              precision    recall  f1-score   support

       happy       0.64      0.88      0.74        32
     neutral       0.54      0.44      0.48        32
         sad       0.61      0.62      0.62        32
    surprise       0.88      0.69      0.77        32

    accuracy                           0.66       128
   macro avg       0.67      0.66      0.65       128
weighted avg       0.67      0.66      0.65       128

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

resnet_model = ResNet50V2(weights="imagenet", include_top=False, input_shape=(img_width, img_height, color_layers))
resnet_model.summary()

Model: "resnet50v2"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer         │ (None, 48, 48, 3) │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_pad           │ (None, 54, 54, 3) │          0 │ input_layer[0][0] │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_conv (Conv2D) │ (None, 24, 24,    │      9,472 │ conv1_pad[0][0]   │
│                     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ pool1_pad           │ (None, 26, 26,    │          0 │ conv1_conv[0][0]  │
│ (ZeroPadding2D)     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ pool1_pool          │ (None, 12, 12,    │          0 │ pool1_pad[0][0]   │
│ (MaxPooling2D)      │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_preac… │ (None, 12, 12,    │        256 │ pool1_pool[0][0]  │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_preac… │ (None, 12, 12,    │          0 │ conv2_block1_pre… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_conv │ (None, 12, 12,    │      4,096 │ conv2_block1_pre… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_bn   │ (None, 12, 12,    │        256 │ conv2_block1_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_relu │ (None, 12, 12,    │          0 │ conv2_block1_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_pad  │ (None, 14, 14,    │          0 │ conv2_block1_1_r… │
│ (ZeroPadding2D)     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_conv │ (None, 12, 12,    │     36,864 │ conv2_block1_2_p… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_bn   │ (None, 12, 12,    │        256 │ conv2_block1_2_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_relu │ (None, 12, 12,    │          0 │ conv2_block1_2_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_0_conv │ (None, 12, 12,    │     16,640 │ conv2_block1_pre… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_3_conv │ (None, 12, 12,    │     16,640 │ conv2_block1_2_r… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_out    │ (None, 12, 12,    │          0 │ conv2_block1_0_c… │
│ (Add)               │ 256)              │            │ conv2_block1_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_preac… │ (None, 12, 12,    │      1,024 │ conv2_block1_out… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_preac… │ (None, 12, 12,    │          0 │ conv2_block2_pre… │
│ (Activation)        │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_conv │ (None, 12, 12,    │     16,384 │ conv2_block2_pre… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_bn   │ (None, 12, 12,    │        256 │ conv2_block2_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_relu │ (None, 12, 12,    │          0 │ conv2_block2_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_pad  │ (None, 14, 14,    │          0 │ conv2_block2_1_r… │
│ (ZeroPadding2D)     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_conv │ (None, 12, 12,    │     36,864 │ conv2_block2_2_p… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_bn   │ (None, 12, 12,    │        256 │ conv2_block2_2_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_relu │ (None, 12, 12,    │          0 │ conv2_block2_2_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_3_conv │ (None, 12, 12,    │     16,640 │ conv2_block2_2_r… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_out    │ (None, 12, 12,    │          0 │ conv2_block1_out… │
│ (Add)               │ 256)              │            │ conv2_block2_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_preac… │ (None, 12, 12,    │      1,024 │ conv2_block2_out… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_preac… │ (None, 12, 12,    │          0 │ conv2_block3_pre… │
│ (Activation)        │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_conv │ (None, 12, 12,    │     16,384 │ conv2_block3_pre… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_bn   │ (None, 12, 12,    │        256 │ conv2_block3_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_relu │ (None, 12, 12,    │          0 │ conv2_block3_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_pad  │ (None, 14, 14,    │          0 │ conv2_block3_1_r… │
│ (ZeroPadding2D)     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_conv │ (None, 6, 6, 64)  │     36,864 │ conv2_block3_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_bn   │ (None, 6, 6, 64)  │        256 │ conv2_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_relu │ (None, 6, 6, 64)  │          0 │ conv2_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ max_pooling2d       │ (None, 6, 6, 256) │          0 │ conv2_block2_out… │
│ (MaxPooling2D)      │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_3_conv │ (None, 6, 6, 256) │     16,640 │ conv2_block3_2_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_out    │ (None, 6, 6, 256) │          0 │ max_pooling2d[0]… │
│ (Add)               │                   │            │ conv2_block3_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_preac… │ (None, 6, 6, 256) │      1,024 │ conv2_block3_out… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_preac… │ (None, 6, 6, 256) │          0 │ conv3_block1_pre… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_conv │ (None, 6, 6, 128) │     32,768 │ conv3_block1_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_bn   │ (None, 6, 6, 128) │        512 │ conv3_block1_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_relu │ (None, 6, 6, 128) │          0 │ conv3_block1_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_pad  │ (None, 8, 8, 128) │          0 │ conv3_block1_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_conv │ (None, 6, 6, 128) │    147,456 │ conv3_block1_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_bn   │ (None, 6, 6, 128) │        512 │ conv3_block1_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_relu │ (None, 6, 6, 128) │          0 │ conv3_block1_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_0_conv │ (None, 6, 6, 512) │    131,584 │ conv3_block1_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_3_conv │ (None, 6, 6, 512) │     66,048 │ conv3_block1_2_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_out    │ (None, 6, 6, 512) │          0 │ conv3_block1_0_c… │
│ (Add)               │                   │            │ conv3_block1_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_preac… │ (None, 6, 6, 512) │      2,048 │ conv3_block1_out… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_preac… │ (None, 6, 6, 512) │          0 │ conv3_block2_pre… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_conv │ (None, 6, 6, 128) │     65,536 │ conv3_block2_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_bn   │ (None, 6, 6, 128) │        512 │ conv3_block2_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_relu │ (None, 6, 6, 128) │          0 │ conv3_block2_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_pad  │ (None, 8, 8, 128) │          0 │ conv3_block2_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_conv │ (None, 6, 6, 128) │    147,456 │ conv3_block2_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_bn   │ (None, 6, 6, 128) │        512 │ conv3_block2_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_relu │ (None, 6, 6, 128) │          0 │ conv3_block2_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_3_conv │ (None, 6, 6, 512) │     66,048 │ conv3_block2_2_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_out    │ (None, 6, 6, 512) │          0 │ conv3_block1_out… │
│ (Add)               │                   │            │ conv3_block2_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_preac… │ (None, 6, 6, 512) │      2,048 │ conv3_block2_out… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_preac… │ (None, 6, 6, 512) │          0 │ conv3_block3_pre… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_conv │ (None, 6, 6, 128) │     65,536 │ conv3_block3_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_bn   │ (None, 6, 6, 128) │        512 │ conv3_block3_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_relu │ (None, 6, 6, 128) │          0 │ conv3_block3_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_pad  │ (None, 8, 8, 128) │          0 │ conv3_block3_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_conv │ (None, 6, 6, 128) │    147,456 │ conv3_block3_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_bn   │ (None, 6, 6, 128) │        512 │ conv3_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_relu │ (None, 6, 6, 128) │          0 │ conv3_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_3_conv │ (None, 6, 6, 512) │     66,048 │ conv3_block3_2_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_out    │ (None, 6, 6, 512) │          0 │ conv3_block2_out… │
│ (Add)               │                   │            │ conv3_block3_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_preac… │ (None, 6, 6, 512) │      2,048 │ conv3_block3_out… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_preac… │ (None, 6, 6, 512) │          0 │ conv3_block4_pre… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_conv │ (None, 6, 6, 128) │     65,536 │ conv3_block4_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_bn   │ (None, 6, 6, 128) │        512 │ conv3_block4_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_relu │ (None, 6, 6, 128) │          0 │ conv3_block4_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_pad  │ (None, 8, 8, 128) │          0 │ conv3_block4_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_conv │ (None, 3, 3, 128) │    147,456 │ conv3_block4_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_bn   │ (None, 3, 3, 128) │        512 │ conv3_block4_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_relu │ (None, 3, 3, 128) │          0 │ conv3_block4_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ max_pooling2d_1     │ (None, 3, 3, 512) │          0 │ conv3_block3_out… │
│ (MaxPooling2D)      │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_3_conv │ (None, 3, 3, 512) │     66,048 │ conv3_block4_2_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_out    │ (None, 3, 3, 512) │          0 │ max_pooling2d_1[… │
│ (Add)               │                   │            │ conv3_block4_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_preac… │ (None, 3, 3, 512) │      2,048 │ conv3_block4_out… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_preac… │ (None, 3, 3, 512) │          0 │ conv4_block1_pre… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_conv │ (None, 3, 3, 256) │    131,072 │ conv4_block1_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block1_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block1_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block1_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_conv │ (None, 3, 3, 256) │    589,824 │ conv4_block1_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block1_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_relu │ (None, 3, 3, 256) │          0 │ conv4_block1_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_0_conv │ (None, 3, 3,      │    525,312 │ conv4_block1_pre… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_3_conv │ (None, 3, 3,      │    263,168 │ conv4_block1_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_out    │ (None, 3, 3,      │          0 │ conv4_block1_0_c… │
│ (Add)               │ 1024)             │            │ conv4_block1_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_preac… │ (None, 3, 3,      │      4,096 │ conv4_block1_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_preac… │ (None, 3, 3,      │          0 │ conv4_block2_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_conv │ (None, 3, 3, 256) │    262,144 │ conv4_block2_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block2_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block2_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block2_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_conv │ (None, 3, 3, 256) │    589,824 │ conv4_block2_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block2_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_relu │ (None, 3, 3, 256) │          0 │ conv4_block2_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_3_conv │ (None, 3, 3,      │    263,168 │ conv4_block2_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_out    │ (None, 3, 3,      │          0 │ conv4_block1_out… │
│ (Add)               │ 1024)             │            │ conv4_block2_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_preac… │ (None, 3, 3,      │      4,096 │ conv4_block2_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_preac… │ (None, 3, 3,      │          0 │ conv4_block3_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_conv │ (None, 3, 3, 256) │    262,144 │ conv4_block3_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block3_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block3_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block3_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_conv │ (None, 3, 3, 256) │    589,824 │ conv4_block3_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_relu │ (None, 3, 3, 256) │          0 │ conv4_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_3_conv │ (None, 3, 3,      │    263,168 │ conv4_block3_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_out    │ (None, 3, 3,      │          0 │ conv4_block2_out… │
│ (Add)               │ 1024)             │            │ conv4_block3_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_preac… │ (None, 3, 3,      │      4,096 │ conv4_block3_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_preac… │ (None, 3, 3,      │          0 │ conv4_block4_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_conv │ (None, 3, 3, 256) │    262,144 │ conv4_block4_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block4_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block4_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block4_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_conv │ (None, 3, 3, 256) │    589,824 │ conv4_block4_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block4_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_relu │ (None, 3, 3, 256) │          0 │ conv4_block4_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_3_conv │ (None, 3, 3,      │    263,168 │ conv4_block4_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_out    │ (None, 3, 3,      │          0 │ conv4_block3_out… │
│ (Add)               │ 1024)             │            │ conv4_block4_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_preac… │ (None, 3, 3,      │      4,096 │ conv4_block4_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_preac… │ (None, 3, 3,      │          0 │ conv4_block5_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_conv │ (None, 3, 3, 256) │    262,144 │ conv4_block5_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block5_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block5_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block5_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_conv │ (None, 3, 3, 256) │    589,824 │ conv4_block5_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block5_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_relu │ (None, 3, 3, 256) │          0 │ conv4_block5_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_3_conv │ (None, 3, 3,      │    263,168 │ conv4_block5_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_out    │ (None, 3, 3,      │          0 │ conv4_block4_out… │
│ (Add)               │ 1024)             │            │ conv4_block5_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_preac… │ (None, 3, 3,      │      4,096 │ conv4_block5_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_preac… │ (None, 3, 3,      │          0 │ conv4_block6_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_conv │ (None, 3, 3, 256) │    262,144 │ conv4_block6_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_bn   │ (None, 3, 3, 256) │      1,024 │ conv4_block6_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_relu │ (None, 3, 3, 256) │          0 │ conv4_block6_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_pad  │ (None, 5, 5, 256) │          0 │ conv4_block6_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_conv │ (None, 2, 2, 256) │    589,824 │ conv4_block6_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_bn   │ (None, 2, 2, 256) │      1,024 │ conv4_block6_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_relu │ (None, 2, 2, 256) │          0 │ conv4_block6_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ max_pooling2d_2     │ (None, 2, 2,      │          0 │ conv4_block5_out… │
│ (MaxPooling2D)      │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_3_conv │ (None, 2, 2,      │    263,168 │ conv4_block6_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_out    │ (None, 2, 2,      │          0 │ max_pooling2d_2[… │
│ (Add)               │ 1024)             │            │ conv4_block6_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_preac… │ (None, 2, 2,      │      4,096 │ conv4_block6_out… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_preac… │ (None, 2, 2,      │          0 │ conv5_block1_pre… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_conv │ (None, 2, 2, 512) │    524,288 │ conv5_block1_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block1_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_relu │ (None, 2, 2, 512) │          0 │ conv5_block1_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_pad  │ (None, 4, 4, 512) │          0 │ conv5_block1_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_conv │ (None, 2, 2, 512) │  2,359,296 │ conv5_block1_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block1_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_relu │ (None, 2, 2, 512) │          0 │ conv5_block1_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_0_conv │ (None, 2, 2,      │  2,099,200 │ conv5_block1_pre… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_3_conv │ (None, 2, 2,      │  1,050,624 │ conv5_block1_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_out    │ (None, 2, 2,      │          0 │ conv5_block1_0_c… │
│ (Add)               │ 2048)             │            │ conv5_block1_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_preac… │ (None, 2, 2,      │      8,192 │ conv5_block1_out… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_preac… │ (None, 2, 2,      │          0 │ conv5_block2_pre… │
│ (Activation)        │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_conv │ (None, 2, 2, 512) │  1,048,576 │ conv5_block2_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block2_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_relu │ (None, 2, 2, 512) │          0 │ conv5_block2_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_pad  │ (None, 4, 4, 512) │          0 │ conv5_block2_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_conv │ (None, 2, 2, 512) │  2,359,296 │ conv5_block2_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block2_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_relu │ (None, 2, 2, 512) │          0 │ conv5_block2_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_3_conv │ (None, 2, 2,      │  1,050,624 │ conv5_block2_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_out    │ (None, 2, 2,      │          0 │ conv5_block1_out… │
│ (Add)               │ 2048)             │            │ conv5_block2_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_preac… │ (None, 2, 2,      │      8,192 │ conv5_block2_out… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_preac… │ (None, 2, 2,      │          0 │ conv5_block3_pre… │
│ (Activation)        │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_conv │ (None, 2, 2, 512) │  1,048,576 │ conv5_block3_pre… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block3_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_relu │ (None, 2, 2, 512) │          0 │ conv5_block3_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_pad  │ (None, 4, 4, 512) │          0 │ conv5_block3_1_r… │
│ (ZeroPadding2D)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_conv │ (None, 2, 2, 512) │  2,359,296 │ conv5_block3_2_p… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_bn   │ (None, 2, 2, 512) │      2,048 │ conv5_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_relu │ (None, 2, 2, 512) │          0 │ conv5_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_3_conv │ (None, 2, 2,      │  1,050,624 │ conv5_block3_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_out    │ (None, 2, 2,      │          0 │ conv5_block2_out… │
│ (Add)               │ 2048)             │            │ conv5_block3_3_c… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ post_bn             │ (None, 2, 2,      │      8,192 │ conv5_block3_out… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ post_relu           │ (None, 2, 2,      │          0 │ post_bn[0][0]     │
│ (Activation)        │ 2048)             │            │                   │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 23,564,800 (89.89 MB)

 Trainable params: 23,519,360 (89.72 MB)

 Non-trainable params: 45,440 (177.50 KB)

# Define a new model that cuts ResNet50V2 at the 'conv3_block4_out' layer
model_output = resnet_model.get_layer("conv3_block4_out").output
cut_model = Model(inputs=resnet_model.input, outputs=model_output)

# Freezing the layers
for layer in resnet_model.layers:
    layer.trainable = False

new_resnet_model = Sequential()
new_resnet_model.add(cut_model)

# Reduces each feature map to a single value by averaging all elements
new_resnet_model.add(GlobalAveragePooling2D())

# Adding full connected layers
new_resnet_model.add(Dense(256, activation="relu"))
# Adding output layer
new_resnet_model.add(Dense(4, activation="softmax"))

# Using Adam Optimizer
optimizer = Adam(learning_rate=0.001)

new_resnet_model.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"])
new_resnet_model.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ functional_1 (Functional)       │ ?                      │     1,453,568 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ global_average_pooling2d        │ ?                      │   0 (unbuilt) │
│ (GlobalAveragePooling2D)        │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 1,453,568 (5.54 MB)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 1,453,568 (5.54 MB)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 20 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=20
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

mc = ModelCheckpoint(
    f"{results_path}/best_model_resnet_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 40 epochs and using validation set
history_resnet = new_resnet_model.fit(
    train_generator_resnet,
    epochs=40,
    validation_data=validation_generator_resnet,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/40

/home/iamtxena/sandbox/mit-ai/my_env/lib/python3.10/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:120: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()

I0000 00:00:1712794843.982897 1499815 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_3697', 4 bytes spill stores, 4 bytes spill loads

I0000 00:00:1712794844.285459 1499816 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_3697', 32 bytes spill stores, 32 bytes spill loads

I0000 00:00:1712794844.474295 1499820 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_3697', 40 bytes spill stores, 40 bytes spill loads

  1/473 ━━━━━━━━━━━━━━━━━━━━ 35:56 5s/step - accuracy: 0.2500 - loss: 1.8365


  4/473 ━━━━━━━━━━━━━━━━━━━━ 8s 18ms/step - accuracy: 0.2051 - loss: 2.3369


  7/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2131 - loss: 2.3027


 10/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2261 - loss: 2.2428


 13/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2377 - loss: 2.1814


 16/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2445 - loss: 2.1280

plt.plot(history_resnet.history["accuracy"])
plt.plot(history_resnet.history["val_accuracy"])
plt.title("ResNet50V2 Model accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator_resnet.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator_resnet.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = new_resnet_model.evaluate(test_generator_resnet, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step - accuracy: 0.6250 - loss: 0.9257


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.5740 - loss: 0.9644

Loss: 0.9330559968948364, Accuracy: 0.578125

pred_probabilities = new_resnet_model.predict(test_generator_resnet, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator_resnet.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("ResNet50V2 Model Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 2s 866ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 1s 5ms/step

              precision    recall  f1-score   support

       happy       0.57      0.62      0.60        32
     neutral       0.40      0.53      0.45        32
         sad       0.71      0.38      0.49        32
    surprise       0.76      0.78      0.77        32

    accuracy                           0.58       128
   macro avg       0.61      0.58      0.58       128
weighted avg       0.61      0.58      0.58       128

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

efficient_model = EfficientNetV2B0(
    weights="imagenet", include_top=False, input_shape=(img_width, img_height, color_layers)
)
# Making all the layers of the efficient_model model non-trainable. i.e. freezing them
for layer in efficient_model.layers:
    layer.trainable = False

efficient_model.summary()

Model: "efficientnetv2-b0"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer         │ (None, 48, 48, 3) │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ rescaling           │ (None, 48, 48, 3) │          0 │ input_layer[0][0] │
│ (Rescaling)         │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ normalization       │ (None, 48, 48, 3) │          0 │ rescaling[0][0]   │
│ (Normalization)     │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ stem_conv (Conv2D)  │ (None, 24, 24,    │        864 │ normalization[0]… │
│                     │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ stem_bn             │ (None, 24, 24,    │        128 │ stem_conv[0][0]   │
│ (BatchNormalizatio… │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ stem_activation     │ (None, 24, 24,    │          0 │ stem_bn[0][0]     │
│ (Activation)        │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block1a_project_co… │ (None, 24, 24,    │      4,608 │ stem_activation[… │
│ (Conv2D)            │ 16)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block1a_project_bn  │ (None, 24, 24,    │         64 │ block1a_project_… │
│ (BatchNormalizatio… │ 16)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block1a_project_ac… │ (None, 24, 24,    │          0 │ block1a_project_… │
│ (Activation)        │ 16)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2a_expand_conv │ (None, 12, 12,    │      9,216 │ block1a_project_… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2a_expand_bn   │ (None, 12, 12,    │        256 │ block2a_expand_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2a_expand_act… │ (None, 12, 12,    │          0 │ block2a_expand_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2a_project_co… │ (None, 12, 12,    │      2,048 │ block2a_expand_a… │
│ (Conv2D)            │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2a_project_bn  │ (None, 12, 12,    │        128 │ block2a_project_… │
│ (BatchNormalizatio… │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_expand_conv │ (None, 12, 12,    │     36,864 │ block2a_project_… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_expand_bn   │ (None, 12, 12,    │        512 │ block2b_expand_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_expand_act… │ (None, 12, 12,    │          0 │ block2b_expand_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_project_co… │ (None, 12, 12,    │      4,096 │ block2b_expand_a… │
│ (Conv2D)            │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_project_bn  │ (None, 12, 12,    │        128 │ block2b_project_… │
│ (BatchNormalizatio… │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_drop        │ (None, 12, 12,    │          0 │ block2b_project_… │
│ (Dropout)           │ 32)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block2b_add (Add)   │ (None, 12, 12,    │          0 │ block2b_drop[0][… │
│                     │ 32)               │            │ block2a_project_… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3a_expand_conv │ (None, 6, 6, 128) │     36,864 │ block2b_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3a_expand_bn   │ (None, 6, 6, 128) │        512 │ block3a_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3a_expand_act… │ (None, 6, 6, 128) │          0 │ block3a_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3a_project_co… │ (None, 6, 6, 48)  │      6,144 │ block3a_expand_a… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3a_project_bn  │ (None, 6, 6, 48)  │        192 │ block3a_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_expand_conv │ (None, 6, 6, 192) │     82,944 │ block3a_project_… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_expand_bn   │ (None, 6, 6, 192) │        768 │ block3b_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_expand_act… │ (None, 6, 6, 192) │          0 │ block3b_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_project_co… │ (None, 6, 6, 48)  │      9,216 │ block3b_expand_a… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_project_bn  │ (None, 6, 6, 48)  │        192 │ block3b_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_drop        │ (None, 6, 6, 48)  │          0 │ block3b_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block3b_add (Add)   │ (None, 6, 6, 48)  │          0 │ block3b_drop[0][… │
│                     │                   │            │ block3a_project_… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_expand_conv │ (None, 6, 6, 192) │      9,216 │ block3b_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_expand_bn   │ (None, 6, 6, 192) │        768 │ block4a_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_expand_act… │ (None, 6, 6, 192) │          0 │ block4a_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_dwconv2     │ (None, 3, 3, 192) │      1,728 │ block4a_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_bn          │ (None, 3, 3, 192) │        768 │ block4a_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_activation  │ (None, 3, 3, 192) │          0 │ block4a_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_se_squeeze  │ (None, 192)       │          0 │ block4a_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_se_reshape  │ (None, 1, 1, 192) │          0 │ block4a_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_se_reduce   │ (None, 1, 1, 12)  │      2,316 │ block4a_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_se_expand   │ (None, 1, 1, 192) │      2,496 │ block4a_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_se_excite   │ (None, 3, 3, 192) │          0 │ block4a_activati… │
│ (Multiply)          │                   │            │ block4a_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_project_co… │ (None, 3, 3, 96)  │     18,432 │ block4a_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4a_project_bn  │ (None, 3, 3, 96)  │        384 │ block4a_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_expand_conv │ (None, 3, 3, 384) │     36,864 │ block4a_project_… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_expand_bn   │ (None, 3, 3, 384) │      1,536 │ block4b_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_expand_act… │ (None, 3, 3, 384) │          0 │ block4b_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_dwconv2     │ (None, 3, 3, 384) │      3,456 │ block4b_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_bn          │ (None, 3, 3, 384) │      1,536 │ block4b_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_activation  │ (None, 3, 3, 384) │          0 │ block4b_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_se_squeeze  │ (None, 384)       │          0 │ block4b_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_se_reshape  │ (None, 1, 1, 384) │          0 │ block4b_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_se_reduce   │ (None, 1, 1, 24)  │      9,240 │ block4b_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_se_expand   │ (None, 1, 1, 384) │      9,600 │ block4b_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_se_excite   │ (None, 3, 3, 384) │          0 │ block4b_activati… │
│ (Multiply)          │                   │            │ block4b_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_project_co… │ (None, 3, 3, 96)  │     36,864 │ block4b_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_project_bn  │ (None, 3, 3, 96)  │        384 │ block4b_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_drop        │ (None, 3, 3, 96)  │          0 │ block4b_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4b_add (Add)   │ (None, 3, 3, 96)  │          0 │ block4b_drop[0][… │
│                     │                   │            │ block4a_project_… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_expand_conv │ (None, 3, 3, 384) │     36,864 │ block4b_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_expand_bn   │ (None, 3, 3, 384) │      1,536 │ block4c_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_expand_act… │ (None, 3, 3, 384) │          0 │ block4c_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_dwconv2     │ (None, 3, 3, 384) │      3,456 │ block4c_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_bn          │ (None, 3, 3, 384) │      1,536 │ block4c_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_activation  │ (None, 3, 3, 384) │          0 │ block4c_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_se_squeeze  │ (None, 384)       │          0 │ block4c_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_se_reshape  │ (None, 1, 1, 384) │          0 │ block4c_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_se_reduce   │ (None, 1, 1, 24)  │      9,240 │ block4c_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_se_expand   │ (None, 1, 1, 384) │      9,600 │ block4c_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_se_excite   │ (None, 3, 3, 384) │          0 │ block4c_activati… │
│ (Multiply)          │                   │            │ block4c_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_project_co… │ (None, 3, 3, 96)  │     36,864 │ block4c_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_project_bn  │ (None, 3, 3, 96)  │        384 │ block4c_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_drop        │ (None, 3, 3, 96)  │          0 │ block4c_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block4c_add (Add)   │ (None, 3, 3, 96)  │          0 │ block4c_drop[0][… │
│                     │                   │            │ block4b_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_expand_conv │ (None, 3, 3, 576) │     55,296 │ block4c_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_expand_bn   │ (None, 3, 3, 576) │      2,304 │ block5a_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_expand_act… │ (None, 3, 3, 576) │          0 │ block5a_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_dwconv2     │ (None, 3, 3, 576) │      5,184 │ block5a_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_bn          │ (None, 3, 3, 576) │      2,304 │ block5a_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_activation  │ (None, 3, 3, 576) │          0 │ block5a_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_se_squeeze  │ (None, 576)       │          0 │ block5a_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_se_reshape  │ (None, 1, 1, 576) │          0 │ block5a_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_se_reduce   │ (None, 1, 1, 24)  │     13,848 │ block5a_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_se_expand   │ (None, 1, 1, 576) │     14,400 │ block5a_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_se_excite   │ (None, 3, 3, 576) │          0 │ block5a_activati… │
│ (Multiply)          │                   │            │ block5a_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_project_co… │ (None, 3, 3, 112) │     64,512 │ block5a_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5a_project_bn  │ (None, 3, 3, 112) │        448 │ block5a_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_expand_conv │ (None, 3, 3, 672) │     75,264 │ block5a_project_… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_expand_bn   │ (None, 3, 3, 672) │      2,688 │ block5b_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_expand_act… │ (None, 3, 3, 672) │          0 │ block5b_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_dwconv2     │ (None, 3, 3, 672) │      6,048 │ block5b_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_bn          │ (None, 3, 3, 672) │      2,688 │ block5b_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_activation  │ (None, 3, 3, 672) │          0 │ block5b_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_se_squeeze  │ (None, 672)       │          0 │ block5b_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_se_reshape  │ (None, 1, 1, 672) │          0 │ block5b_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_se_reduce   │ (None, 1, 1, 28)  │     18,844 │ block5b_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_se_expand   │ (None, 1, 1, 672) │     19,488 │ block5b_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_se_excite   │ (None, 3, 3, 672) │          0 │ block5b_activati… │
│ (Multiply)          │                   │            │ block5b_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_project_co… │ (None, 3, 3, 112) │     75,264 │ block5b_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_project_bn  │ (None, 3, 3, 112) │        448 │ block5b_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_drop        │ (None, 3, 3, 112) │          0 │ block5b_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5b_add (Add)   │ (None, 3, 3, 112) │          0 │ block5b_drop[0][… │
│                     │                   │            │ block5a_project_… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_expand_conv │ (None, 3, 3, 672) │     75,264 │ block5b_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_expand_bn   │ (None, 3, 3, 672) │      2,688 │ block5c_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_expand_act… │ (None, 3, 3, 672) │          0 │ block5c_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_dwconv2     │ (None, 3, 3, 672) │      6,048 │ block5c_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_bn          │ (None, 3, 3, 672) │      2,688 │ block5c_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_activation  │ (None, 3, 3, 672) │          0 │ block5c_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_se_squeeze  │ (None, 672)       │          0 │ block5c_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_se_reshape  │ (None, 1, 1, 672) │          0 │ block5c_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_se_reduce   │ (None, 1, 1, 28)  │     18,844 │ block5c_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_se_expand   │ (None, 1, 1, 672) │     19,488 │ block5c_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_se_excite   │ (None, 3, 3, 672) │          0 │ block5c_activati… │
│ (Multiply)          │                   │            │ block5c_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_project_co… │ (None, 3, 3, 112) │     75,264 │ block5c_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_project_bn  │ (None, 3, 3, 112) │        448 │ block5c_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_drop        │ (None, 3, 3, 112) │          0 │ block5c_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5c_add (Add)   │ (None, 3, 3, 112) │          0 │ block5c_drop[0][… │
│                     │                   │            │ block5b_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_expand_conv │ (None, 3, 3, 672) │     75,264 │ block5c_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_expand_bn   │ (None, 3, 3, 672) │      2,688 │ block5d_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_expand_act… │ (None, 3, 3, 672) │          0 │ block5d_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_dwconv2     │ (None, 3, 3, 672) │      6,048 │ block5d_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_bn          │ (None, 3, 3, 672) │      2,688 │ block5d_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_activation  │ (None, 3, 3, 672) │          0 │ block5d_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_se_squeeze  │ (None, 672)       │          0 │ block5d_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_se_reshape  │ (None, 1, 1, 672) │          0 │ block5d_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_se_reduce   │ (None, 1, 1, 28)  │     18,844 │ block5d_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_se_expand   │ (None, 1, 1, 672) │     19,488 │ block5d_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_se_excite   │ (None, 3, 3, 672) │          0 │ block5d_activati… │
│ (Multiply)          │                   │            │ block5d_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_project_co… │ (None, 3, 3, 112) │     75,264 │ block5d_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_project_bn  │ (None, 3, 3, 112) │        448 │ block5d_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_drop        │ (None, 3, 3, 112) │          0 │ block5d_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5d_add (Add)   │ (None, 3, 3, 112) │          0 │ block5d_drop[0][… │
│                     │                   │            │ block5c_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_expand_conv │ (None, 3, 3, 672) │     75,264 │ block5d_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_expand_bn   │ (None, 3, 3, 672) │      2,688 │ block5e_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_expand_act… │ (None, 3, 3, 672) │          0 │ block5e_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_dwconv2     │ (None, 3, 3, 672) │      6,048 │ block5e_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_bn          │ (None, 3, 3, 672) │      2,688 │ block5e_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_activation  │ (None, 3, 3, 672) │          0 │ block5e_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_se_squeeze  │ (None, 672)       │          0 │ block5e_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_se_reshape  │ (None, 1, 1, 672) │          0 │ block5e_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_se_reduce   │ (None, 1, 1, 28)  │     18,844 │ block5e_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_se_expand   │ (None, 1, 1, 672) │     19,488 │ block5e_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_se_excite   │ (None, 3, 3, 672) │          0 │ block5e_activati… │
│ (Multiply)          │                   │            │ block5e_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_project_co… │ (None, 3, 3, 112) │     75,264 │ block5e_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_project_bn  │ (None, 3, 3, 112) │        448 │ block5e_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_drop        │ (None, 3, 3, 112) │          0 │ block5e_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block5e_add (Add)   │ (None, 3, 3, 112) │          0 │ block5e_drop[0][… │
│                     │                   │            │ block5d_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_expand_conv │ (None, 3, 3, 672) │     75,264 │ block5e_add[0][0] │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_expand_bn   │ (None, 3, 3, 672) │      2,688 │ block6a_expand_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_expand_act… │ (None, 3, 3, 672) │          0 │ block6a_expand_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_dwconv2     │ (None, 2, 2, 672) │      6,048 │ block6a_expand_a… │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_bn          │ (None, 2, 2, 672) │      2,688 │ block6a_dwconv2[… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_activation  │ (None, 2, 2, 672) │          0 │ block6a_bn[0][0]  │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_se_squeeze  │ (None, 672)       │          0 │ block6a_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_se_reshape  │ (None, 1, 1, 672) │          0 │ block6a_se_squee… │
│ (Reshape)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_se_reduce   │ (None, 1, 1, 28)  │     18,844 │ block6a_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_se_expand   │ (None, 1, 1, 672) │     19,488 │ block6a_se_reduc… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_se_excite   │ (None, 2, 2, 672) │          0 │ block6a_activati… │
│ (Multiply)          │                   │            │ block6a_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_project_co… │ (None, 2, 2, 192) │    129,024 │ block6a_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6a_project_bn  │ (None, 2, 2, 192) │        768 │ block6a_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_expand_conv │ (None, 2, 2,      │    221,184 │ block6a_project_… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_expand_bn   │ (None, 2, 2,      │      4,608 │ block6b_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_expand_act… │ (None, 2, 2,      │          0 │ block6b_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_dwconv2     │ (None, 2, 2,      │     10,368 │ block6b_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_bn          │ (None, 2, 2,      │      4,608 │ block6b_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_activation  │ (None, 2, 2,      │          0 │ block6b_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_se_squeeze  │ (None, 1152)      │          0 │ block6b_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_se_reshape  │ (None, 1, 1,      │          0 │ block6b_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6b_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_se_expand   │ (None, 1, 1,      │     56,448 │ block6b_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_se_excite   │ (None, 2, 2,      │          0 │ block6b_activati… │
│ (Multiply)          │ 1152)             │            │ block6b_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_project_co… │ (None, 2, 2, 192) │    221,184 │ block6b_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_project_bn  │ (None, 2, 2, 192) │        768 │ block6b_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_drop        │ (None, 2, 2, 192) │          0 │ block6b_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6b_add (Add)   │ (None, 2, 2, 192) │          0 │ block6b_drop[0][… │
│                     │                   │            │ block6a_project_… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_expand_conv │ (None, 2, 2,      │    221,184 │ block6b_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_expand_bn   │ (None, 2, 2,      │      4,608 │ block6c_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_expand_act… │ (None, 2, 2,      │          0 │ block6c_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_dwconv2     │ (None, 2, 2,      │     10,368 │ block6c_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_bn          │ (None, 2, 2,      │      4,608 │ block6c_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_activation  │ (None, 2, 2,      │          0 │ block6c_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_se_squeeze  │ (None, 1152)      │          0 │ block6c_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_se_reshape  │ (None, 1, 1,      │          0 │ block6c_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6c_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_se_expand   │ (None, 1, 1,      │     56,448 │ block6c_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_se_excite   │ (None, 2, 2,      │          0 │ block6c_activati… │
│ (Multiply)          │ 1152)             │            │ block6c_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_project_co… │ (None, 2, 2, 192) │    221,184 │ block6c_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_project_bn  │ (None, 2, 2, 192) │        768 │ block6c_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_drop        │ (None, 2, 2, 192) │          0 │ block6c_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6c_add (Add)   │ (None, 2, 2, 192) │          0 │ block6c_drop[0][… │
│                     │                   │            │ block6b_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_expand_conv │ (None, 2, 2,      │    221,184 │ block6c_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_expand_bn   │ (None, 2, 2,      │      4,608 │ block6d_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_expand_act… │ (None, 2, 2,      │          0 │ block6d_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_dwconv2     │ (None, 2, 2,      │     10,368 │ block6d_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_bn          │ (None, 2, 2,      │      4,608 │ block6d_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_activation  │ (None, 2, 2,      │          0 │ block6d_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_se_squeeze  │ (None, 1152)      │          0 │ block6d_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_se_reshape  │ (None, 1, 1,      │          0 │ block6d_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6d_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_se_expand   │ (None, 1, 1,      │     56,448 │ block6d_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_se_excite   │ (None, 2, 2,      │          0 │ block6d_activati… │
│ (Multiply)          │ 1152)             │            │ block6d_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_project_co… │ (None, 2, 2, 192) │    221,184 │ block6d_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_project_bn  │ (None, 2, 2, 192) │        768 │ block6d_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_drop        │ (None, 2, 2, 192) │          0 │ block6d_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6d_add (Add)   │ (None, 2, 2, 192) │          0 │ block6d_drop[0][… │
│                     │                   │            │ block6c_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_expand_conv │ (None, 2, 2,      │    221,184 │ block6d_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_expand_bn   │ (None, 2, 2,      │      4,608 │ block6e_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_expand_act… │ (None, 2, 2,      │          0 │ block6e_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_dwconv2     │ (None, 2, 2,      │     10,368 │ block6e_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_bn          │ (None, 2, 2,      │      4,608 │ block6e_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_activation  │ (None, 2, 2,      │          0 │ block6e_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_se_squeeze  │ (None, 1152)      │          0 │ block6e_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_se_reshape  │ (None, 1, 1,      │          0 │ block6e_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6e_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_se_expand   │ (None, 1, 1,      │     56,448 │ block6e_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_se_excite   │ (None, 2, 2,      │          0 │ block6e_activati… │
│ (Multiply)          │ 1152)             │            │ block6e_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_project_co… │ (None, 2, 2, 192) │    221,184 │ block6e_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_project_bn  │ (None, 2, 2, 192) │        768 │ block6e_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_drop        │ (None, 2, 2, 192) │          0 │ block6e_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6e_add (Add)   │ (None, 2, 2, 192) │          0 │ block6e_drop[0][… │
│                     │                   │            │ block6d_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_expand_conv │ (None, 2, 2,      │    221,184 │ block6e_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_expand_bn   │ (None, 2, 2,      │      4,608 │ block6f_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_expand_act… │ (None, 2, 2,      │          0 │ block6f_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_dwconv2     │ (None, 2, 2,      │     10,368 │ block6f_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_bn          │ (None, 2, 2,      │      4,608 │ block6f_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_activation  │ (None, 2, 2,      │          0 │ block6f_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_se_squeeze  │ (None, 1152)      │          0 │ block6f_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_se_reshape  │ (None, 1, 1,      │          0 │ block6f_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6f_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_se_expand   │ (None, 1, 1,      │     56,448 │ block6f_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_se_excite   │ (None, 2, 2,      │          0 │ block6f_activati… │
│ (Multiply)          │ 1152)             │            │ block6f_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_project_co… │ (None, 2, 2, 192) │    221,184 │ block6f_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_project_bn  │ (None, 2, 2, 192) │        768 │ block6f_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_drop        │ (None, 2, 2, 192) │          0 │ block6f_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6f_add (Add)   │ (None, 2, 2, 192) │          0 │ block6f_drop[0][… │
│                     │                   │            │ block6e_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_expand_conv │ (None, 2, 2,      │    221,184 │ block6f_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_expand_bn   │ (None, 2, 2,      │      4,608 │ block6g_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_expand_act… │ (None, 2, 2,      │          0 │ block6g_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_dwconv2     │ (None, 2, 2,      │     10,368 │ block6g_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_bn          │ (None, 2, 2,      │      4,608 │ block6g_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_activation  │ (None, 2, 2,      │          0 │ block6g_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_se_squeeze  │ (None, 1152)      │          0 │ block6g_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_se_reshape  │ (None, 1, 1,      │          0 │ block6g_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6g_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_se_expand   │ (None, 1, 1,      │     56,448 │ block6g_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_se_excite   │ (None, 2, 2,      │          0 │ block6g_activati… │
│ (Multiply)          │ 1152)             │            │ block6g_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_project_co… │ (None, 2, 2, 192) │    221,184 │ block6g_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_project_bn  │ (None, 2, 2, 192) │        768 │ block6g_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_drop        │ (None, 2, 2, 192) │          0 │ block6g_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6g_add (Add)   │ (None, 2, 2, 192) │          0 │ block6g_drop[0][… │
│                     │                   │            │ block6f_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_expand_conv │ (None, 2, 2,      │    221,184 │ block6g_add[0][0] │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_expand_bn   │ (None, 2, 2,      │      4,608 │ block6h_expand_c… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_expand_act… │ (None, 2, 2,      │          0 │ block6h_expand_b… │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_dwconv2     │ (None, 2, 2,      │     10,368 │ block6h_expand_a… │
│ (DepthwiseConv2D)   │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_bn          │ (None, 2, 2,      │      4,608 │ block6h_dwconv2[… │
│ (BatchNormalizatio… │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_activation  │ (None, 2, 2,      │          0 │ block6h_bn[0][0]  │
│ (Activation)        │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_se_squeeze  │ (None, 1152)      │          0 │ block6h_activati… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_se_reshape  │ (None, 1, 1,      │          0 │ block6h_se_squee… │
│ (Reshape)           │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_se_reduce   │ (None, 1, 1, 48)  │     55,344 │ block6h_se_resha… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_se_expand   │ (None, 1, 1,      │     56,448 │ block6h_se_reduc… │
│ (Conv2D)            │ 1152)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_se_excite   │ (None, 2, 2,      │          0 │ block6h_activati… │
│ (Multiply)          │ 1152)             │            │ block6h_se_expan… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_project_co… │ (None, 2, 2, 192) │    221,184 │ block6h_se_excit… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_project_bn  │ (None, 2, 2, 192) │        768 │ block6h_project_… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_drop        │ (None, 2, 2, 192) │          0 │ block6h_project_… │
│ (Dropout)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ block6h_add (Add)   │ (None, 2, 2, 192) │          0 │ block6h_drop[0][… │
│                     │                   │            │ block6g_add[0][0] │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ top_conv (Conv2D)   │ (None, 2, 2,      │    245,760 │ block6h_add[0][0] │
│                     │ 1280)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ top_bn              │ (None, 2, 2,      │      5,120 │ top_conv[0][0]    │
│ (BatchNormalizatio… │ 1280)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ top_activation      │ (None, 2, 2,      │          0 │ top_bn[0][0]      │
│ (Activation)        │ 1280)             │            │                   │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 5,919,312 (22.58 MB)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 5,919,312 (22.58 MB)

new_efficient_model = Sequential()
new_efficient_model.add(efficient_model)

# Reduces each feature map to a single value by averaging all elements
new_efficient_model.add(GlobalAveragePooling2D())

# Adding full connected layers
new_efficient_model.add(Dense(256, activation="relu"))
new_efficient_model.add(Dense(128, activation="relu"))

# Output Layer
new_efficient_model.add(Dense(4, activation="softmax"))

# Using Adam Optimizer
optimizer = Adam(learning_rate=0.01)

new_efficient_model.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"])
new_efficient_model.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ efficientnetv2-b0 (Functional)  │ ?                      │     5,919,312 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ global_average_pooling2d        │ ?                      │   0 (unbuilt) │
│ (GlobalAveragePooling2D)        │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 5,919,312 (22.58 MB)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 5,919,312 (22.58 MB)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 20 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=20
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

# Define the saving the best model callback
mc = ModelCheckpoint(
    f"{results_path}/best_model_efficient_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 40 epochs and using validation set
history_efficient = new_efficient_model.fit(
    train_generator_efficientnet,
    epochs=40,
    validation_data=validation_generator_efficientnet,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/40

/home/iamtxena/sandbox/mit-ai/my_env/lib/python3.10/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:120: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()

I0000 00:00:1712795242.855338 1509123 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11837', 32 bytes spill stores, 32 bytes spill loads

I0000 00:00:1712795242.912001 1509129 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11837', 40 bytes spill stores, 40 bytes spill loads

I0000 00:00:1712795243.210471 1509126 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11548', 32 bytes spill stores, 32 bytes spill loads

I0000 00:00:1712795243.323623 1509129 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11837', 4 bytes spill stores, 4 bytes spill loads

I0000 00:00:1712795243.486678 1509130 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11858', 4 bytes spill stores, 4 bytes spill loads

I0000 00:00:1712795243.743913 1509132 asm_compiler.cc:369] ptxas warning : Registers are spilled to local memory in function 'triton_gemm_dot_11548', 244 bytes spill stores, 244 bytes spill loads

  1/473 ━━━━━━━━━━━━━━━━━━━━ 2:55:47 22s/step - accuracy: 0.4062 - loss: 1.3525


  5/473 ━━━━━━━━━━━━━━━━━━━━ 7s 17ms/step - accuracy: 0.3559 - loss: 2.2122


  8/473 ━━━━━━━━━━━━━━━━━━━━ 8s 18ms/step - accuracy: 0.3463 - loss: 2.2160


 11/473 ━━━━━━━━━━━━━━━━━━━━ 8s 18ms/step - accuracy: 0.3443 - loss: 2.1482

plt.plot(history_efficient.history["accuracy"])
plt.plot(history_efficient.history["val_accuracy"])
plt.title("EfficientNet Model accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator_efficientnet.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator_efficientnet.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = new_efficient_model.evaluate(test_generator_efficientnet, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 27ms/step - accuracy: 0.8125 - loss: 0.6692


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - accuracy: 0.6635 - loss: 0.8448

Loss: 0.8933918476104736, Accuracy: 0.625

pred_probabilities = new_efficient_model.predict(test_generator_efficientnet, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator_efficientnet.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("EfficientNet Model Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 7s 3s/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 3s 5ms/step

              precision    recall  f1-score   support

       happy       0.53      0.81      0.64        32
     neutral       0.50      0.47      0.48        32
         sad       0.78      0.56      0.65        32
    surprise       0.81      0.66      0.72        32

    accuracy                           0.62       128
   macro avg       0.66      0.62      0.63       128
weighted avg       0.66      0.62      0.63       128

# Set this to 'grayscale' as the images are in grayscale
color_mode = "grayscale"
color_layers = 1
# As we have checked, all images are 48x48, we will set the img_width and img_height to 48
img_width, img_height = 48, 48
# A batch size of 32 is appropriate for this dataset provide to provide a good balance
# between the model's ability to generalize (avoid overfitting) and computational efficiency.
batch_size = 32

# Training Data Augmentation
train_datagen = ImageDataGenerator(
    rescale=1.0 / 255,  # Normalize pixel values to [0,1]
    horizontal_flip=True,  # Faces are symmetric; flipping can simulate looking from another direction
    brightness_range=(0.5, 1.5),  # Randomly adjust brightness to simulate different lighting conditions
    shear_range=0.3,  # Shear transformations for perspective changes
    rotation_range=20,  # Slight rotation to introduce variability without distorting emotion features
    width_shift_range=0.1,  # Slight horizontal shifts to simulate off-center faces
    height_shift_range=0.1,  # Slight vertical shifts to account for different heights/angles
    zoom_range=0.1,  # Small zoom in/out to simulate closer or further away faces
)

# Validation and Testing Data should not be augmented!
validation_datagen = ImageDataGenerator(rescale=1.0 / 255)
test_datagen = ImageDataGenerator(rescale=1.0 / 255)

# Creating train_dir, validation_dir, and test_dir with the structure of DATADIR and SUBDIRS
train_dir = os.path.join(DATADIR, SUBDIRS_DICT["train"])
validation_dir = os.path.join(DATADIR, SUBDIRS_DICT["validation"])
test_dir = os.path.join(DATADIR, SUBDIRS_DICT["test"])

# Train Generator
train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
)

# Validation Generator
validation_generator = validation_datagen.flow_from_directory(
    validation_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for evaluation
)

# Testing Generator
test_generator = test_datagen.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode=color_mode,  # Set to 'grayscale'
    class_mode="categorical",
    shuffle=False,  # shuffle=False to keep data in order for testing
)

Found 15109 images belonging to 4 classes.

Found 4977 images belonging to 4 classes.

Found 128 images belonging to 4 classes.

backend.clear_session()

# Fixing the seed for random number generators so that we can ensure we receive the same output everytime
np.random.seed(42)

random.seed(42)
tf.random.set_seed(42)

# Initializing a sequential model
model_complex = Sequential()

model_complex.add(Input(shape=(img_width, img_height, color_layers)))

# First Convolutional Block
model_complex.add(Conv2D(64, kernel_size=2, padding="same"))
model_complex.add(BatchNormalization())
model_complex.add(LeakyReLU(negative_slope=0.1))
model_complex.add(MaxPooling2D(pool_size=2))

# Second Convolutional Block
model_complex.add(Conv2D(128, kernel_size=2, padding="same"))
model_complex.add(BatchNormalization())
model_complex.add(LeakyReLU(negative_slope=0.1))
model_complex.add(MaxPooling2D(pool_size=2))

# Third Convolutional Block
model_complex.add(Conv2D(256, kernel_size=2, padding="same"))
model_complex.add(BatchNormalization())
model_complex.add(LeakyReLU(negative_slope=0.1))
model_complex.add(MaxPooling2D(pool_size=2))

# Fourth Convolutional Block
model_complex.add(Conv2D(512, kernel_size=2, padding="same"))
model_complex.add(BatchNormalization())
model_complex.add(LeakyReLU(negative_slope=0.1))
model_complex.add(MaxPooling2D(pool_size=2))

# Fifth Convolutional Block
model_complex.add(Conv2D(128, kernel_size=2, padding="same"))
model_complex.add(BatchNormalization())
model_complex.add(LeakyReLU(negative_slope=0.1))
model_complex.add(MaxPooling2D(pool_size=2))

# Flatten the output of the conv layers to feed into the dense layers
model_complex.add(Flatten())

model_complex.add(Dense(512, activation="relu"))
model_complex.add(Dense(128, activation="relu"))
model_complex.add(Dense(64))
model_complex.add(BatchNormalization())
model_complex.add(ReLU())  # Using ReLU after batch normalization
model_complex.add(Dense(4, activation="softmax"))

# Using RMSProp Optimizer
optimizer = RMSprop(learning_rate=0.01)

model_complex.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"])
model_complex.summary()

Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ conv2d (Conv2D)                 │ (None, 48, 48, 64)     │           320 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization             │ (None, 48, 48, 64)     │           256 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu (LeakyReLU)         │ (None, 48, 48, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d (MaxPooling2D)    │ (None, 24, 24, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_1 (Conv2D)               │ (None, 24, 24, 128)    │        32,896 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_1           │ (None, 24, 24, 128)    │           512 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_1 (LeakyReLU)       │ (None, 24, 24, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_1 (MaxPooling2D)  │ (None, 12, 12, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_2 (Conv2D)               │ (None, 12, 12, 256)    │       131,328 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_2           │ (None, 12, 12, 256)    │         1,024 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_2 (LeakyReLU)       │ (None, 12, 12, 256)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_2 (MaxPooling2D)  │ (None, 6, 6, 256)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_3 (Conv2D)               │ (None, 6, 6, 512)      │       524,800 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_3           │ (None, 6, 6, 512)      │         2,048 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_3 (LeakyReLU)       │ (None, 6, 6, 512)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_3 (MaxPooling2D)  │ (None, 3, 3, 512)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_4 (Conv2D)               │ (None, 3, 3, 128)      │       262,272 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_4           │ (None, 3, 3, 128)      │           512 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_4 (LeakyReLU)       │ (None, 3, 3, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_4 (MaxPooling2D)  │ (None, 1, 1, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten (Flatten)               │ (None, 128)            │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ (None, 512)            │        66,048 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 128)            │        65,664 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ (None, 64)             │         8,256 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_5           │ (None, 64)             │           256 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu (ReLU)                    │ (None, 64)             │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_3 (Dense)                 │ (None, 4)              │           260 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 1,096,452 (4.18 MB)

 Trainable params: 1,094,148 (4.17 MB)

 Non-trainable params: 2,304 (9.00 KB)

# Get the current time
current_time = datetime.now().strftime("%Y%m%d-%H%M%S")

# Set up Early Stopping with a patience 7 but acting after at least 30 epochs
delayed_early_stopping = DelayedEarlyStopping(
    monitor="val_loss", patience=7, verbose=1, restore_best_weights=True, start_epoch=30
)

# Define the learning rate scheduler callback
reduce_lr = ReduceLROnPlateau(monitor="val_loss", factor=0.2, patience=5, min_lr=0.00001, verbose=1)

# Define the saving the best model callback
mc = ModelCheckpoint(
    f"{results_path}/best_model_complex_{current_time}.keras",
    monitor="val_accuracy",
    mode="max",
    verbose=1,
    save_best_only=True,
)

# Fitting the model with 60 epochs and using validation set
history_complex = model_complex.fit(
    train_generator,
    epochs=60,
    validation_data=validation_generator,
    callbacks=[reduce_lr, mc, delayed_early_stopping],
)

Epoch 1/60

/home/iamtxena/sandbox/mit-ai/my_env/lib/python3.10/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:120: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()

  1/473 ━━━━━━━━━━━━━━━━━━━━ 25:59 3s/step - accuracy: 0.1875 - loss: 1.7487


  5/473 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.2273 - loss: 1.7262


  7/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2378 - loss: 1.6855


 10/473 ━━━━━━━━━━━━━━━━━━━━ 9s 21ms/step - accuracy: 0.2517 - loss: 1.6414


 13/473 ━━━━━━━━━━━━━━━━━━━━ 9s 20ms/step - accuracy: 0.2598 - loss: 1.6109


 17/473 ━━━━━━━━━━━━━━━━━━━━ 8s 19ms/step - accuracy: 0.2619 - loss: 1.5875


 22/473 ━━━━━━━━━━━━━━━━━━━━ 7s 17ms/step - accuracy: 0.2643 - loss: 1.5641


 27/473 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.2666 - loss: 1.5457

plt.plot(history_complex.history["accuracy"])
plt.plot(history_complex.history["val_accuracy"])
plt.title("Complex CNN Model accuracy")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(["train", "validation"], loc="upper left")
plt.show()

# Calculate the number of steps for the entire test set to be processed
test_steps = test_generator.samples // batch_size

# If the number of samples isn't a multiple of the batch size,
# you have one more batch with the remaining samples
if test_generator.samples % batch_size > 0:
    test_steps += 1

# Evaluating the model on the test set
evaluation_results = model_complex.evaluate(test_generator, steps=test_steps)
print(f"Loss: {evaluation_results[0]}, Accuracy: {evaluation_results[1]}")

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 24ms/step - accuracy: 0.8750 - loss: 0.3037


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - accuracy: 0.8281 - loss: 0.4269

Loss: 0.46025848388671875, Accuracy: 0.8125

pred_probabilities = model_complex.predict(test_generator, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("Complex CNN Model Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 1s 453ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step

              precision    recall  f1-score   support

       happy       0.93      0.88      0.90        32
     neutral       0.69      0.84      0.76        32
         sad       0.74      0.62      0.68        32
    surprise       0.91      0.91      0.91        32

    accuracy                           0.81       128
   macro avg       0.82      0.81      0.81       128
weighted avg       0.82      0.81      0.81       128

pred_probabilities = model_complex.predict(test_generator, steps=test_steps)
pred = np.argmax(pred_probabilities, axis=1)

# Getting the true labels from the generator
y_true = test_generator.classes

# Printing the classification report with actual emotion labels
print(classification_report(y_true, pred, target_names=CATEGORIES))

# Plotting the heatmap using confusion matrix
cm = confusion_matrix(y_true, pred)
plt.figure(figsize=(8, 5))
sns.heatmap(cm, annot=True, fmt=".0f", xticklabels=CATEGORIES, yticklabels=CATEGORIES)
plt.title("Best Model Confusion Matrix")
plt.ylabel("Actual")
plt.xlabel("Predicted")
plt.show()

1/4 ━━━━━━━━━━━━━━━━━━━━ 0s 24ms/step


4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step

              precision    recall  f1-score   support

       happy       0.93      0.88      0.90        32
     neutral       0.69      0.84      0.76        32
         sad       0.74      0.62      0.68        32
    surprise       0.91      0.91      0.91        32

    accuracy                           0.81       128
   macro avg       0.82      0.81      0.81       128
weighted avg       0.82      0.81      0.81       128

Facial Emotion Detection¶

Executive Summary¶

Context¶

Objective¶

Key Questions¶

Problem Formulation¶

About the dataset¶

Importing the Libraries¶

Let us load and unzip the data¶

Preparing the Data¶

Visualizing our Classes¶

Happy¶

Sad¶

Neutral¶

Surprised¶

Checking Distribution of Classes¶

Creating our Data Loaders¶

Model Building¶

Creating the Base Neural Network¶

Model 1 Architecture:¶

Compiling and Training the Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the Model on the Test Set¶

Plotting Confusion Matrix¶

Creating the second Convolutional Neural Network¶

Model 2 Architecture:¶

Compiling and Training the Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the Model on the Test Set¶

Plotting Confusion Matrix¶

Think About It:¶

Transfer Learning Architectures¶

Creating our Data Loaders for Transfer Learning Architectures¶

VGG16 Model¶

Importing the VGG16 Architecture¶

Model Building¶

Compiling and Training the VGG16 Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the VGG16 model¶

Plotting Confusion Matrix¶

ResNet V2 Model¶

Model Building¶

Compiling and Training the Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the ResNet Model¶

Plotting Confusion Matrix¶

EfficientNet Model¶

Model Building¶

Compiling and Training the Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the EfficientnetNet Model¶

Plotting the confusion matrix¶

Building a Complex Neural Network Architecture¶

Creating our Data Loaders¶

Model Building¶

Compiling and Training the Model¶

Plotting the Training and Validation Accuracies¶

Evaluating the Model on Test Set¶

Plotting Confusion Matrix¶

Plotting the Confusion Matrix for the chosen final model¶

Conclusion: The custom complex CNN model, with an accuracy of 81%, outperformed transfer learning architectures, providing an efficient and accurate solution for facial emotion recognition in grayscale images.¶

Insights¶

Refined insights:¶

Comparison of various techniques and their relative performance:¶

Proposal for the final solution design:¶

Recommendations for Implementation¶

Key Recommendations¶

Actionables for Stakeholders¶

Expected Benefit/Cost¶

Key Risks and Challenges¶

Further Analysis and Associated Problems¶

Appendix¶

Project Code Repository Overview¶

Final Notebook and Batch Testing¶